Regional feedbacks among fire, climate, and tropical deforestationWilliam A. Hoffmann1Departamento de Engenharia Florestal, Universidade de Brası´lia, Brası´lia, BrazilWilfrid SchroederInstituto Brasileiro do Meio Ambiente e dos Recursos Naturais Renova´veis, Brası´lia, BrazilRobert B. JacksonDepartment of Biology and Nicholas School of the Environment and Earth Sciences, Duke University, Durham,North Carolina, USAReceived 11 February 2003; revised 30 June 2003; accepted 31 July 2003; published 4 December 2003.[1] Numerous studies with general circulation models suggest that tropical deforestationcan result in regional-scale climate change; namely, increased air temperature and windspeed and reduced precipitation and relative humidity. To quantify how this climatechange should affect fire risk, we used the National Center for Atmospheric Research(NCAR) CCM3.2 general circulation model and remote sensing to estimate the effect oftropical deforestation on fire risk through the McArthur forest fire danger index (FFDI).Deforestation reduced precipitation and relative humidity and increased wind speed inthe Amazon, Congo, and Indonesia/New Guinea. FFDI increased by 41, 56, and 58% inthese three regions, respectively, primarily owing to higher wind speeds and reducedprecipitation. Actual fire occurrence in the Amazon, as determined from NOAA-12images, was strongly correlated with the FFDI calculated from meteorological data (P 0.0001). Using the observed relationship between FFDI and fire occurrence, we estimateincreases in fire frequency of 44, 80, and 123%, in the Amazon, Congo, and Indonesia,respectively, with deforestation. In all three regions the largest relative increases in firerisk occurred in the more humid areas with the lowest original fire risk.INDEX TERMS:0315 Atmospheric Composition and Structure: Biosphere/atmosphere interactions; 3360 Meteorology andAtmospheric Dynamics: Remote sensing; 3210 Mathematical Geophysics: Modeling; 3322 Meteorology andAtmospheric Dynamics: Land/atmosphere interactions; KEYWORDS: tropical forest, fire, GCM modeling,remote sensing, fire meteorology, climate changeCitation: Hoffmann, W. A., W. Schroeder, and R. B. Jackson, Regional feedbacks among fire, climate, and tropical deforestation,J. Geophys. Res., 108(D23), 4721, doi:10.1029/2003JD003494, 2003.1. Introduction[2] Uncontrolled fire is a principal factor contributing tothe degradation of deforested and selectively logged tropicalforests. Thinning or removal of the forest canopy permitsgreater insolation at the soil surface, which dries fuel,increases air temperature and reduces relative humidity nearthe soil [Uhl and Kauffman, 1990]. So although undisturbedtropical forest is not typically flammab le, even duringmoderate drought, selectively logged forest and areascleared for pasture are prone to burning [Holdsworth andUhl, 1997; Uhl and Kauffman, 1990]. Large areas of pastureand selectively logged forest burn annually [Cochrane etal., 1999; Nepstad et al., 1999]. This burning furtherreduces tree cover and prevents tree regeneration, resultingin a positive feedback at the local scale [Cochrane et al.,1999; Cochrane and Schulze , 1999; Nepstad et al., 2001].[3] In addition to local changes in microclimate, regionalclimate change resulting from large-scale deforestationshould contribute further to this vegetation-climate feedback[Laurance and Williamson, 2001; Hoffmann and Jackson,2000; Hoffmann et al., 2002; Nepstad et al., 2001]. Simu-lations with general circulation models (GCMs) haverepeatedly demonstrated that changes in albedo, roughnesslength, leaf-area index and rooting depth caused by tropicaldeforestation reduce precipitation and relative humidity andincrease surface temperature and wind speed [Dickinsonand Kennedy, 1992; Hahmann and Dickinson, 1997;Henderson-Sellers et al., 1993; Nobre et al., 1991; Polcherand Laval, 1994; Sud et al. , 1996; Zeng et al., 1996; Zhanget al., 1996]. All four of these climatic changes shouldincrease fire risk [Hoffmann et al., 2002; Noble et al., 1980].[4] Although these changes are often described asregional effects of deforestation, such GCM results combineboth local and regional effects [Salati and Nobre, 1991].Distinguishing between regional and local climate change isJOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. D23, 4721, doi:10.1029/2003JD003494, 20031Now at Department of Botany, North Carolina State University,Raleigh, North Carolina, USA.Copyright 2003 by the American Geophysical Union.0148-0227/03/2003JD003494$09.00ACL 4 - 1important because the two are likely to have very differentimpacts on fire occurrence. If the changes are entirely local,the increased fire risk would be limited only to disturbedsites and would be relatively unaffected by the regionalextent of deforestation. In contrast, a regional component tothis feedback could extend the increased fire risk to undis-turbed areas and would depend more strongly on the scaleof deforestation.[5] Here we focus on the regional effects of tropicaldeforestation on fire occurrence, using GCM simulationsand remote sensing to answer the following questions: Whatare the predicted changes in fire risk and fire occurrence dueto tropical deforestation? What are the relative contributionsof precipitation, relative humidity, wind speed, and temper-ature to this overall change in fire risk? How does theincrease in fire risk respond to the scale of deforestation?2. Model and Simulations[6] Simulations were run using the NCAR CommunityClimate Model (CCM3.2) with a spatial resolution ofapproximately 2.8  2.8 of Earth’s surface (T42 spectraltruncation) and 18 vertical levels. A detailed model descrip-tion is available from Kiehl et al. [1998]. CCM3.2 is coupledwith the NCAR Land Surface Model (LSM) described byBonan [1996]. LSM simulates the fluxes of momentum,radiation, latent heat and sensible heat between the land andthe atmosphere. An analysis of the control climatology ofCCM3.2/LSM is presented by Bonan [1998].[7] Within LSM, each grid cell of the vegetated surface ofEarth is assigned one of 28 vegetation types. Each vegetationtype is composed of one or more plant types and/or bare soil.When vegetation is composed of more than one plant type,surface variables as well as fluxes of water, energy, andmomentum are averaged over separate vegetation subgridcells, each occupied by the respective plant type. A total of12 plant types are represented in LSM, differing in leaf area,stem